How does exposure to violence affect school delay and academic motivation for adolescents living in

socioeconomically disadvantaged communities in South Africa?

Authors: Herrero Romero, R., Hall, J., Cluver, L., Meinck, F., Hinde, E

Abstract

To date, little is known about the effects of violence on the educational outcomes of adolescents

in disadvantaged communities in South Africa. In response, self-report data was collected from a

socioeconomically disadvantaged sample of 503 adolescents aged 10 to 18 participating in a

child abuse prevention trial in the Eastern Cape. Adolescents were purposively selected in the

trial. This study applies Latent Profile Analysis (LPA) to examine relationships between past-

month exposure to violence, school delay and academic motivation. 93.8% of adolescents in the

sample experienced “poly-violence” – exposure to at least two forms of violence in the past

month. Results identified two distinct profiles in the socioeconomically disadvantaged sample:

Profile 1, adolescents exposed to more frequent “poly-violence”, and Profile 2, adolescents

exposed to less frequent “poly-violence”. Being exposed to more frequent “poly-violence” was

associated with greater risk of school delay – based on age-appropriate grade in South Africa.

However, being exposed to more frequent “poly-violence” was not associated with lower

academic motivation – adolescents showed high rates of wanting to achieve. Our findings

suggest that exposure to more frequent “poly-violence” increases risk of school delay among

adolescents from disadvantaged communities, while not affecting their academic motivation.

Thus, whilst adolescents maintained aspirations and goals to do well at school, exposure to high

frequency of violence impacted their capacity to fulfill these aims.

Keywords

Violence, academic motivation, school delay, adolescence, disadvantaged communities, South

Africa, LPA

How does exposure to violence affect school delay and academic motivation for adolescents living in socioeconomically disadvantaged communities in South Africa?

Adolescents’ educational outcomes in disadvantaged communities in South Africa

School delay and low completion rates are common among disadvantaged youths in

South Africa (Lam, Ardington, & Leibbrandt, 2011). Nationally representative, South African

evidence has shown that adolescents from socioeconomically disadvantaged families and

communities attending disadvantaged schools in disadvantaged areas are less likely to perform

well and adequately progress in school (Branson, Hofmeyr, & Lam, 2014; Spaull, 2013). For

instance, by grade 9 –age 15-, the average student from a poorly-resourced school is likely to

perform at a level commensurate with an average student at least three years their junior but

attending a more “functional” school – usually one charging fees- (Spaull, 2013). Despite this,

past studies have suggested that disadvantaged students in South Africa have high hopes and

aspirations in terms of education (Bray, Gooskens, Kahn, Moses, & Seekings, 2010; Mzolo,

2013; Watson, Mcmahon, Foxcroft, & Els, 2010) as well as high motivation to keep trying in

school (Strassburg, Meny-Gibert, & Russell, 2010; Ward, Martin, Theron, & Distiller, 2007).

Despite their socioeconomic circumstances, disadvantaged adolescents in South Africa persevere

in order to remain in school and complete basic education (Grade 9). This results in a large

number of over-aged students due to low performance and consequent grade repetition. Among

them, only a very small proportion complete non-compulsory education (Grade 12) (Spaull,

2015).

South African adolescents’ exposure to violence

In South Africa, high rates of exposure to violence were reported in the first nationally

representative survey on childhood victimization. Amongst adolescents aged 15 to 17, 34.4%

reported having experienced physical abuse by caregivers, 21.3% experienced neglect, 16%

experienced emotional abuse, and 23.1% witnessed domestic violence –physical, emotional or

sexual violence between other household members such as caregivers and siblings- in their

lifetime (Ward, Artz, Burton, & Leoschut, 2015).

In addition to exposure to violence at home, adolescents in South Africa can also be

exposed to high levels of violence perpetrated by peers and adults outside the home. For

example, harsh discipline in school by teachers, bullying practices by peers, or assaults in the

community are common (Ward et al., 2015, 2007). At the school level, 19.7% of all South

African adolescents report persistent bullying (Ward et al., 2007), while 66.9% of all high

school students in the Eastern Cape have experienced corporal punishment (Burton & Leoschut,

2013b) and 50% reported having witnessed a physical fight in their community in their own

lives (Ward et al., 2015). Overall, 64% of all adolescents aged 15 to 17 in South Africa

experienced “Lifetime Poly-victimization” - numerous victimizations across different contexts

throughout their lives (Leoschut & Kafaar, 2017). “Lifetime Poly-victimization” is more

frequent amongst socioeconomically disadvantaged adolescents due to chronic poverty,

unemployment, parental stress, and household overcrowding (Burton & Leoschut, 2013a;

Leoschut & Kafaar, 2017; Meinck, Cluver, & Boyes, 2013, 2015; Seedat, Niekerk, Suffla, &

Ratele, 2009; Ward et al., 2015).

Exposure to violence and poor educational outcomes

There is strong global and South African evidence demonstrating the impacts of exposure

to violence on children’s mental and physical health (Barbarin, Richter, & Wet, 2001; Meinck,

Cluver, Orkin, et al., 2016; WHO, 2014). However fewer studies have focused on the

relationships between children’s exposure to violence and educational outcomes.

While evidence on the relationship between violence and adolescents’ educational

outcomes in Sub-Saharan Africa is scarce, more research can be found in other regions. For

instance, a recent review found 16 cross-sectional retrospective studies from the US, Canada,

Israel and the UK, showing that childhood maltreatment within the family was associated with

later poor academic performance (Romano, Babchishin, Marquis, & Fre, 2015). This review

found that physical abuse, sexual abuse and neglect were independently associated with

enrolment in lower grades, higher grade repetition, and academic underachievement.

Furthermore, the negative impact of maltreatment within the family on academic outcomes was

more pronounced for children with multiple maltreatment experiences, as well as children

exposed to specific types of maltreatment such as neglect (Romano et al., 2015). Similarly, a

cross-sectional study with adolescent students from an urban, ethnically diverse school in the US

showed that youth with multiple victimizations outside the home (bullying, conventional crime

and sexual victimization) experienced more psychological distress and earned lower grades than

their peers (Holt, Finkelhor, & Kantor, 2007). Likewise, a cross-sectional study with young

adolescents from urban areas in Jamaica found that school-based peer violence, physical

punishment at school, and community violence were independently associated with poor school

achievement. In addition, children experiencing higher levels of violence had the poorest

academic achievement compared to other children, suggesting a dose-response effect (Baker-

henningham, Meeks-gardner, Chang, & Walker, 2009).

In South Africa, one longitudinal and two cross-sectional studies exist investigating the

impact of harsh parenting, plus exposure to domestic and community violence on the educational

outcomes of children and adolescents (Barbarin et al., 2001; Pieterse, 2015; Sherr et al., 2016).

In a cross-sectional study of 4,747 youths aged 14–22 in Cape Town, being physically hit hard

on a regular basis was associated with an increased probability of school dropout and a reduction

in numeracy test scores (Pieterse, 2015). Similar results were obtained in a study of 626 six-

year-old children in South Africa. Here, family violence was inversely associated with academic

motivation, and community violence was not found to affect academic functioning (Barbarin et

al., 2001). Also, results from a one-year longitudinal study of children aged 4-13 showed that

children exposed to emotional abuse by their parents at baseline were ten times less likely to still

be enrolled in school for the one year follow-up (Sherr et al., 2016). In the same study, harsh

parenting was also associated with poor grade progression.

The three studies described above all used ‘variable-based analyses’ (Petrenko, Friend,

Garrido, Taussig, & Culhane, 2013) to investigate single relationships between specific forms of

violence and negative educational outcomes among young children (Barbarin et al., 2001; Sherr

et al., 2016) and adolescents (Pieterse, 2015; Sherr et al., 2016) in South Africa. However none

of these studies analyzed the impact of exposure to multiple forms of violence on educational

outcomes among adolescents. Additionally, contrary to research in other regions none of these

South African studies found significant associations between exposure to community violence

and educational outcomes.

Measuring exposure to multiple forms of violence

Exposure to multiple forms of violence has been operationalized using two different

approaches: variable-centered (i.e. cumulative risk indices) versus person-centered approaches

(i.e. Latent Class and Latent Profile Analysis). By creating a continuous measure - an index

summating dichotomized items (at risk-thresholds) from several sub-scales-, researchers have

considered endorsement to any dichotomized item in the index as a presence of violence

victimization (Charak et al., 2016; Finkelhor, Ormrod, & Turner, 2007). Using this construct of

violence victimization, researchers have classified adults and children into different groups or

classes. These classes range from “none”, “least victimized–exposed” or “one type of violence

victimization” to “highly poly-victimized–exposed” and/or, “more than three violence

victimizations” (Charak et al., 2016; Finkelhor et al., 2007). Thus, for these researchers, “poly-

victimization” indicates that a person has experienced at least four different types of violence

victimizations in the past year (Charak et al., 2016; Finkelhor et al., 2007). This approach to

“poly-victimization” has two limitations: first, these models are not able to consider repeated

victimization or the frequency of episodes of violence; second, they give the same importance to

all different types of experiences of violence.

A second method is to use a person-centered alternative that considers multiple violence

experiences based on various violence scale scores, such as Latent Class Analysis (LCA) or

Latent Profile Analysis (LPA) (Holt et al., 2007). This approach allows comparisons between

different levels of exposure to violence by asking participants how often they have experienced

different types of victimization in the past month. For instance, by using Latent Profile Analysis

(LPA), participants can be classified in groups based on their victimization profiles (for

example, high neglect/low physical and emotional abuse profile versus high neglect, physical

and emotional abuse profile versus high sexual and emotional abuse profile). These profiles can

then be compared to each other based on their mean scores (Holt et al., 2007). Compared to

variable-centered techniques, LPA avoids the use of arbitrary cut-off points on underlying

dimensions and makes it possible to retain the continuous nature of frequency ratings of

exposure to violence (Pears, Kim, & Fisher, 2008). Thus, the strength of this method, which has

been used in other studies on child maltreatment, is its ability to identify various sub-groups of

children with similar victimization profiles, based on the frequency and type of violence

exposure (Pears et al., 2008).

Current Study

This study examines the relationship between South African adolescents’ exposure to

multiple types of violence with subsequent academic motivation and school delay. Exposure to

violence is investigated across multiple settings -home, school and community-, and by different

perpetrators –caregivers, peers and other adults. The term “poly-violence” is used to indicate

exposure to more than one of the following six forms of violence: domestic violence between

household members, physical abuse by caregivers, emotional abuse by caregivers, school

violence, community violence (victimization) and community violence (witnessing). Six total

scores based on standardized sub-scales made the six ‘forms’ of violence. Thus, the current

study takes into consideration not only the number of forms of violence that adolescents

experience but also the frequency of exposure to each form of violence –based on the mean

scores- (Holt et al., 2007). A prospective approach -5 to 9 months- is applied to a sample of

adolescents from socioeconomically disadvantaged communities. The current study answers the

following research question: Is exposure to violence related to school delay and academic

motivation in adolescents from socioeconomically disadvantaged communities in South Africa?

Methods

Study setting

Baseline and post-test survey data were collected from participants of the Sinovuyo Teen

Study (STS); a pragmatic cluster randomized-controlled trial (Anonymous 2016). This study

evaluated the effectiveness of the Sinovuyo Teen program, a parenting intervention primarily

aimed at reducing harsh and abusive parenting. The trial took place in 32 rural and 8 peri-urban

disadvantaged communities in the Eastern Cape, South Africa. Communities had high rates of

unemployment, HIV prevalence and poor infrastructure. Baseline data collection took place

from May until September 2015, while the post-test intervention surveys (hereafter, ‘follow-up’)

were conducted from January to July 2016. Additional information regarding the purposive

sampling procedure and overall study design can be found in the publication of the trial protocol

(Anonymous 2016).

Screening and recruitment procedures

Participants were Xhosa adolescents aged 10 to 18 years and their primary caregivers

living in 32 rural and 8 peri-urban disadvantaged communities in the Eastern Cape Province of

South Africa. Due to large numbers of orphans in South Africa and the common geographical

mobility of children across extended family households (Yaw, Heaton, & Kalule-Sabiti, 2007),

there were no requirements for a biological relationship between caregiver and adolescent. Very

low-income communities in the Eastern Cape were selected and local schools, traditional

chieftains, and social services in contact with vulnerable families were approached. Participants

were recruited through local social workers, school teachers and community guides who

identified families experiencing arguments in the household and adolescents with social and

emotional problems. Some of the families also self-referred. Families who reported having

experienced arguments within the previous month were then given the opportunity to participate.

Adolescents aged 10 to 18 and sleeping in the same dwelling as their caregiver for at

least four nights a week were eligible to participate in the overall study (n=552). Of those, four

adolescents did not complete baseline interviews, and thus they were not included in this

analysis. Finally, only adolescents who did not subsequently drop out from the trial (n=22), and

had no missing data on educational outcomes in the follow-up survey (n=23) were included in

the current analysis (total included n=503). Adolescents who had missing data on educational

outcomes in the post-test survey reported being out-of-school when the post-test interviews took

place. Due to the skip patterns designed into the Computer-Assisted Self-Interview Software

(CASI) applied in the study, as well as regular data quality and validation checks, missing data

was 0% on all the variables used in this study except for one. Perceived school violence had

<1% of missing data, corresponding to five adolescents who were out of school at baseline.

There were no baseline differences between adolescents lost and retained at follow-up in the

overall study (see Supplement Table 1 in Anonymous et al. 2018).

Data collection

Tablet-based questionnaires translated into Xhosa were used within a Computer-Assisted

Personal Interviewing (CAPI) process. Local, trained research assistants interviewed participants

at their homes or in other settings, such as schools. School level characteristics were also

collected from 69 schools which accounted for 94% of the sample’s adolescents. Administrative

data for those schools where data was not collected directly (n=14) was retrieved from the online

master lists of South African schools (Department of Basic Education, 2016; Van Wyk, 2015).

Ethics

Ethical approval was obtained from Research Ethics Committees at the University of

Oxford, the University of Cape Town, as well as from the South African Departments of Social

Development, and of Basic Education. Both adolescents and caregivers needed to give informed

consent to participate. Confidentiality was maintained, except if participants were at risk of

significant harm or requested assistance. A total of 33 adolescents and their families were

referred to social and health services due to family issues, sexual abuse and suicide risk. A small

‘thank you pack’ of stationery was given to all adolescents for each completed interview as well

as refreshments and certificates of participation. Adolescents were given the choice to answer

abuse questions using Audio Computer-Assisted Self-Interview Software (ACASI). School

principals gave informed consent for the school to participate in the study.

Measures

Composite variables for educational outcomes (measured at follow-up) and for violence

exposure (measured at baseline) are summarized in Table 1.

[Please insert Table 1 about here]

Academic Motivation and School Delay (measured at follow-up)

School delay was assessed using a continuous scale based on grade appropriateness by

age in South Africa. The South African Schools Act of 1996 specifies that the age of admission

to Grade 1 is the year in which a child turns 7 (Republic of South Africa, 1996). However, a

decision made by the Constitutional Court in 2004, resulted in the school-going age for Grade 1

being changed to age 5 if a child turns 6 on or before 30 June in the Grade 1 year (Department of

Basic Education, 2010). Furthermore, many parents decide that their child should start school

earlier. Subject to the availability of places, a learner may be admitted to Grade 1 at a younger

age if it is in the child’s best interests (UNESCO International Bureau of Education, 2006).

Thus, school delay calculations used a conservative definition of age norm per grade in the

current investigation (e.g. grade 1 + 6 = age 7; grade 9 + 6 = age 15; grade 12 + 6 = age 18).

Adolescents who were enrolled in the age-appropriate grade scored 0 in the School Delay scale.

A positive score indicated that the adolescent had dropped behind the South African age-

appropriate grade, while a negative score indicated a higher number of grades ahead.

(Low) Academic motivation was measured using four adapted items from the

standardized ‘Academic Motivation Scale’ of the SAHA study (Ruchkin, Schwab-Stone, &

Vermeiren, 2004); the original scale was validated in the US (α =.66) and the adapted items were

used in previous research with adolescents from socioeconomically disadvantaged communities

in Cape Town (Ward et al., 2007). In the current investigation, these items were answered using

a scale from 0 to 9 with lower scores indicating higher levels of academic motivation. The items

were adapted to fit the South African school context. For instance, we used, “It is important to

me to do well in school this year” instead of “It is important to me to get at least a B average this

year”.

Measures recording exposure to violence (measured at baseline)

Domestic violence between household members, adolescent physical abuse by caregivers

and adolescent emotional abuse by caregivers were measured using a culturally adapted version

of the ISPCAN Child Abuse Screening Tool (ICAST); Children’s Version ICAST-C (Runyan et

al., 2009), the ICAST-TRIAL. The ICAST-TRIAL measured the prevalence of exposure to

violence in the past month. Response codes range from ‘0’ (Never) to ‘8’ (8 times or more)

(Meinck et al., n.d.). Higher scale scores indicated a greater frequency of violence experiences.

Three additional items from the Alabama Parenting Questionnaire (APQ) Corporal Punishment

sub-scale (Frick, 1991), also adapted to measure the past month, were added to the ICAST-

TRIAL to composite the teen physical abuse by caregivers measure. APQ items’ response

options range on a 5-point Likert scale from ‘Never’ to ‘Always’. APQ response values were

multiplied by two in order to match the eight response options of the ICAST-C child physical

abuse scale before summing. In the current analysis, neglect was not considered a form of

violence as such, but as the parents’ failure to care for a child even if able to do so (Ward et al.,

2015). Furthermore, in the context of high levels of deprivation, neglect is difficult to distinguish

from poverty (Ward et al., 2015). Thus, neglect was not included in the analysis. Internal

consistency (Cronbach’s alpha) of the violence subscales were .67 (Domestic violence between

household members, 2 items), .81 (Adolescent emotional abuse by caregivers, 4 items) and.88

(Adolescent physical abuse by caregivers, 4 items). Further information on the psychometric

properties of the ISPCAN Child Abuse Screening Tool for use in Trials among South African

Adolescents can be found in Meinck et al. (under review).

Perceived school violence was measured using five items from the Student Survey

Physical and Emotional Safety sub-scale used in the UNICEF Safe and Caring Child-Friendly

Schools Study (UNICEF, 2009). This sub-scale measures how physically and emotionally safe

students felt in school in the past term and includes items asking about bullying, and crime in

both the school grounds and surroundings. The response code ranged on a 4-point Likert scale

from ‘Definitely not true’ to ‘Definitely true’ with higher scores showing higher levels of

unsafety. The psychometric properties of the original student-reported measure can be found in a

cross-national study using data from 68 schools across the Philippines, Nicaragua and South

Africa – including schools in in extremely difficult socio-economic conditions in the Eastern

Cape province (Godfrey et al., 2012). In the global evaluation, internal consistency of the

original Physical and Emotional Safety sub-scale across all countries was .74. Internal

consistency of the perceived school violence scale in the current investigation was .94.

Witnessing community violence and community violence victimization were measured

using risks identified in the child version of the standardized Exposure to Violence Scale from

the Social and Health Assessment (SAHA) study (Ruchkin et al., 2004; Ward et al., 2007). This

sub-scale was also validated in the US and has since been used in research with adolescents from

socioeconomically disadvantaged communities in Cape Town (Ward, Martin, Theron, &

Distiller, 2007; Ruchkin et al., 2004). The scale consisted of two five-item sub-scales assessing

incidents of children witnessing and experiencing community violence in the past month. The

response code ranged from “Never” to “More than 5 times”. In the current investigation, internal

consistency (Cronbach’s alpha) of the SAHA Exposure to Violence subscales were .78

(Witnessing community violence) and .63 (Community violence victimization).

Probabilistic predictors of violence (measured at baseline)

Adolescents’ age (years) and gender (male/female) were measured via self-report.

Household poverty was measured using a continuous measure (8 items) indicating access to the

top eight socially-perceived necessities for children, as identified in the SA Social Attitudes

Survey (Pillay, Roberts, & Rule, 2006; Wright, 2008). These include items such as, “enough

clothes to keep you warm and dry’ and ‘a visit to the doctor when someone was ill”. At follow-

up, adolescents’ exposure to violence and adolescents’ educational outcomes may have been

influenced by their participation in the STS (intervention versus control). In order to control for

this participation, trial arm was introduced as a covariate.

Data Analysis

All analyses were conducted using the Mplus v7.0 software (Muthén & Muthén, 2010).

The effect of families being nested within communities (though not schools) was statistically

controlled for via the specification of clustered standard errors. This decision to control for the

effect of nesting within one of the two higher levels (cross-classified multilevel data) was made

on the basis of two sets of descriptive statistics: Design effects and Intra Class Correlations

(ICCs; see Table 2).

[Please insert Table 2 about here]

Putting the statistics within Table 2 into context, within studies considering educational

outcomes, a threshold ICC value of 10-15% is considered medium to high effect (Hox, 2010).

Thus, the ICCs shown in Table 2 indicated that a small proportion of variance in School Delay

was due to differences between communities (variance estimate=0.15, p<0.05, ICC=8.4%).

However, no significant results were found due to differences between schools on either School

Delay or Academic Motivation. This might be explained by the very similar type of schools that

the adolescents attended (see schooling characteristics, Table 3). Clustered standard errors were

calculated using the TYPE=COMPLEX feature in Mplus (‘disaggregated multilevel modelling’;

Muthén & Muthén, 2010).Latent Profile Analysis (LPA) was then used to explore distinct

patterns of exposure to violence in the sample. LPA is a person-centered type of analysis; a

mixture model which identifies unobserved homogeneous groups or subpopulations from a

heterogeneous population; these homogeneous groups are based on mean-level response

patterns across multiple continuous risk indicators (Masyn, 2013). In contrast to variable-

centered models, person-centered models allow researchers to identify differences in the

exposure to multiple and co-occurring risks and outcomes of interest between sub-groups of

children (Masten, 2014). Furthermore, Latent Class and Latent Profile Analysis are increasingly

used to measure the interplay of different risk factors and multiple forms of exposure to violence

(Charak et al., 2016; Clarke et al., 2016; Lanza, Tan, & Bray, 2013; Petrenko et al., 2013).

In the current investigation, incidences of domestic violence between household

members, adolescent physical abuse by caregivers, and adolescent emotional abuse by

caregivers were measured using count data (number of times a violent episode occurred in the

past month). On the other hand, perceived school violence, witnessing community violence and

community violence victimization were measured using the Likert response format (ordinal data

to indicate frequency of exposure to violence). In the current study, all item scores within each

violence sub-scale were added. LPA was therefore applied, treating ordinal data as ordinal

approximations to continuous variables. Thus, the total number of categories for perceived

school violence, witnessing community violence and community violence victimization were

twenty, twenty-five and twenty-five respectively. This is a common practice used by researchers

within surveys, as ordinal variables with five or more categories can often be treated as

continuous variables without detriment to the analysis (Johnson & Creech, 1998; Norman,

2010; Zumbo & Zimmerman, 1993).

The following six violence risk indicators were included in the LPA: domestic violence

between household members, adolescent physical abuse by caregivers, adolescent emotional

abuse by caregivers, perceived school violence, community violence (victimization) and

community violence (witnessing). Maximum-likelihood estimation was used to estimate missing

data on the latent profile risk indicators. In order to compare educational outcomes across

profiles, a one-step ‘distal-as-indicator’ model was applied (Lanza et al., 2013) with six

exposure-to-violence risk indicators measured at baseline and the two educational outcomes

measured at follow-up. With a view to identifying factors associated with individual profile

membership, probabilistic predictors (age, gender, poverty and trial arm) were included in the

model via the post-hoc AUXILIARY option in Mplus (Muthén & Muthén, 2010). This option

applies multinomial logistic regressions with the latent profiles as dependent variables using

posterior probabilities (Muthen & Muthen, 2012).

After conducting a two-profile LPA, successive LPAs estimating three through five

profiles were then estimated in order to fully test for the possibility of multiple homogenous

groups/ latent profiles (Nylund & Muthén, 2007). Three types of indices of relative model fit

were computed to inform the number of latent profiles that were deemed to exist (Masyn, 2013;

Nylund & Muthén, 2007). First, the Lo-Mendell-Rubin Likelihood Ratio Test (LMR-LRT) of

goodness of fit, pointing at the ‘best’ model with the smallest number of classes that is not

significantly improved by the addition of another class (Masyn, 2013) Second, information

criteria Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were

examined, where lower values indicate a better model, although sometimes minimum values are

not reached (Masyn, 2013; Nylund & Muthén, 2007). Finally, classification quality was assessed

using relative entropy, with higher values being superior (Masyn, 2013).

Results

Sample characteristics

Table 3 displays the characteristics of the sample at baseline, showing high levels of

socioeconomic disadvantage. Three quarters of adolescents lived in households where at least

three basic necessities were not covered (75.9%), more than 60% of the adolescents’ households

had no tap water and approximately 70% lived with no employed adults at home.

[Please insert Table 3 about here]

All adolescents attended state schools and the vast majority received free meals at school

(96.2%). Adolescents mainly attended poorly-resourced schools in low income areas (85.5%).

While around 80% of adolescents lived in rural areas, only 66.6% attended schools in rural areas

and the rest of the sample (33.4%) attended schools in urban areas, meaning that 14.4% travelled

to urban areas in order to attend school. 37.6% of adolescents were in grades lower than those

considered appropriate for their age. Overall, adolescents showed a high level of academic

motivation and the mean academic motivation score was 28.35 (scale ranged from 0 to 36).

Interestingly, 13% of adolescents in the sample were enrolled two years above the age-

appropriate grade. This is consistent with South African statistics of early first-time enrolment in

poorer areas, compared to more affluent provinces (Department of Basic Education, 2010,

2013). For instance, in 2009, the poorer provinces of Limpopo and the Eastern Cape showed

particularly high proportions of 5-year-olds attending educational institutions, compared to more

affluent provinces such as the Western Cape and Gauteng (Department of Basic Education,

2010). The issue of early school enrolment in disadvantaged areas in South Africa has been

associated with the national roll-out of public early childhood education (Early Childhood

Development and Foundation Phase), as well as the school feeding program (National School

Nutrition Programme; UNESCO International Bureau of Education, 2006; Department of

Education, 2010). Thus, due to early entrance to school in South Africa, it is possible to find

children enrolled in a grade one or two years higher than what is age appropriate, especially in

primary school if children did not repeat grades.

Overall, adolescents in the sample experienced high rates of violence in the month prior

to the baseline survey: 51.1% witnessed some form of domestic violence, 63.2% experienced

some form of physical violence by their caregivers, 76.1% experienced some form of emotional

violence by their caregiver, 88.3% felt unsafe in school, and 71.8% had witnessed some form of

violence in the community. Exposure to community violence (victimization) was considerably

lower in the study sample (32.4%), compared to other forms of violence. Only four adolescents

reported never experiencing any form of violence in the past month, while 93.84% of

adolescents were exposed to “poly-violence”.

LPA Model

The results of the LPAs with 2-5 latent profiles are displayed in Table 4. In terms of

information and classification criteria, models with a greater number of profiles had the smaller

values of AIC and BIC, and higher values of entropy. An inspection of the interpretability of the

models showed that model 3 identified a non-poly-violence profile with 6.96% of adolescents,

plausibly corresponding to the combined proportions of adolescents exposed to one type of

violence (6.2%) and adolescents not exposed to any type of violence (0.8%). However, LMR-

LRT was not significant for models 3, 4 and 5, indicating that none of these significantly

improved upon the initial two-profile model (Masyn, 2013). By contrast, LMR-LRT was

significant for the 2 latent profile solution, indicating that a model with two latent profiles

significantly improved a model with one latent profile. Overall, these results indicated that the

two-profile solution was the best-fitting model - for this sample - and the classification of

adolescents in the two-profile solution was meaningful and had demonstrated greatest parsimony

(Lanza & Rhoades, 2014).

[Please insert Table 4 about here]

Profiles of exposure to violence. The LPA procedure suggested that there were two

heterogeneous profiles of adolescents - differentiated according to the frequency of “poly-

violence” that they were exposed to. Figure 1 compares patterns of exposure to each type of

violence for the two profiles identified. Table 5 shows the differences in the estimated means of

exposure to violence between the two profiles. Adolescents in Profile 1, adolescents exposed to

more frequent “poly-violence” (n=60, 11.9%), were characterized by higher mean scores of

domestic violence (p<0.001), physical (p<0.001) and emotional (p<0.001) abuse at home as well

as higher scores of witnessing community violence (p=0.001), compared to adolescents in

Profile 2. By contrast, adolescents in Profile 2, adolescents exposed to less frequent “poly-

violence” (n=443, 88.1%), were characterized by lower mean scores of violence at home and

witnessing community violence, compared to adolescents in Profile 1. The two profiles showed

very similar scores of perceived school violence (p= 0.500) and community violence

victimization (p=0.060).

[Please insert Table 5 and Figure 1 about here]

Table 6 compares the number of adolescents exposed to each form of violence in the past

month for the two profiles identified. 100% of adolescents in Profile 1 and 92.9% of adolescents

in Profile 2 experienced “poly-violence” in the past month, while 28.3% adolescents in Profile 1

and 9% adolescents in Profile 2 were exposed to all forms of violence in the past month.

[Please insert Table 6 about here]

Educational outcomes and their association with poly-violence. Adolescents in Profile 1

at baseline were less likely to be in the appropriate grade for their age at follow-up (thus

experiencing school delay), compared to adolescents in Profile 2 (see Table 6). Hence, more

frequent exposure to “poly-violence” was associated with school delay in these adolescents.

School delay outcome was measured using a continuous variable (with negative numbers

indicating number of grades ahead). Results therefore also show that children in grades higher

than those age-appropriate were more likely to be in profile 1, and thus, these children were less-

frequently exposed to poly-violence. Academic motivation was not significantly different across

profiles.

Predictors of violence. Multinomial results showed a significant relationship between the

adolescents’ age and membership of each of the exposure-to-violence profiles (see Table 7).

Thus, older adolescents were more likely to be in Profile 1 and to have experienced higher

frequency of poly-violence, compared to younger adolescents (p =0.020). Neither adolescent

gender, family poverty, nor trial arm were significantly associated with profile membership.

[Please insert Table 7]

Discussion

This paper contributes to our knowledge on the lives of South African adolescents living

in poverty by considering the relationship that exists between their exposure to “poly-violence”

and subsequent measures of academic motivation and school delay. To our knowledge, this is

the first study that has examined the impact of exposure to multiple forms of violence in the

home, school and community on the educational outcomes of these adolescents (including

younger and older adolescents). Furthermore, this is the first study to do so from a prospective

approach, and the first that has applied a person-centered approach to measure any group of

adolescents’ exposure to violence in South Africa (the identification of heterogeneous groups).

Exposure to Violence

Overall rates of exposure to any form of violence in the home and in the community were

high in the study sample, compared to national estimates of exposure to violence (Akmatov,

2011; Ward et al., 2015). This was to be expected, given the purposive and socioeconomically

disadvantaged status of the sample. Perceived school violence was particularly high in our

sample, compared to national estimates (Department of Basic Education, 2013). In total, more

than 80% of adolescents in the sample felt unsafe in school. Adolescents in the sample attended

mainly socioeconomically disadvantaged schools in rural areas and townships. Previous research

has demonstrated that endemic crime in socioeconomically disadvantaged communities and a

lack of appropriate supervision during school breaks –often due to a lack of resources- are

important factors contributing to the poor safety in public schools (Burton & Leoschut, 2013a;

Masitsa, 2011).

Despite overall high rates of exposure to “poly-violence", our results strongly confirmed

heterogeneity within our sample. All sampled adolescents lived in socioeconomically

disadvantaged communities characterized by a lack of infrastructure, and high rates of crime and

unemployment. Yet results suggest that not all adolescents were equally exposed to violence and

abuse. Two distinct groups of adolescents can be observed in relation to exposure to violence in

socioeconomically disadvantaged communities in South Africa: 1) adolescents that are exposed

to more frequent “poly-violence” in the home and community, and 2) adolescents that are

exposed to less frequent “poly-violence”, compared to adolescents in group 1. Addressing

exposure to “poly-violence” in socioeconomically disadvantaged communities in South Africa is

particularly important. Research from other countries has shown that adolescents exposed to

multiple forms of violence are more likely to experience more psychological distress and lower

academic performance than those with minimal victimization or those exposed to only one

specific form of violence (Holt et al., 2007).

Furthermore, profile characteristics in the study sample indicated that different

frequencies of exposure to “poly-violence”, both direct and indirect, were present in adolescents

in socioeconomically disadvantaged communities in South Africa. LPA analyses did not yield

profiles characterized by distinct forms of exposure to violence (i.e. exposed to physical versus

emotional abuse, exposed to school versus community violence etc.). However, two clear

profiles characterized by different frequency of exposure to “poly-violence” were identified.

This finding is in line with previous research on children’s exposure to violence in South Africa

and other Sub-Saharan countries, pointing at high rates of children’s multiple victimization

(Clarke et al., 2016; Meinck, Cluver, Boyes, & Loening-voysey, 2016; Ward et al., 2015, 2007).

Thus, results suggested that two groups of adolescents exposed to multiple types of violence can

be found in socioeconomically disadvantaged communities in South Africa: first, those that are

frequently exposed to direct (physical and emotional abuse) and indirect violence at home

(domestic violence), as well as indirect violence in the community; second, those adolescents

that are similarly exposed to family and community violence but in a less frequent manner.

Particularly targeting adolescents exposed to more frequent exposure to “poly-violence” in

socioeconomically disadvantaged communities in South Africa is important: previous research

from different countries has shown that chronic abuse -recurrent incidents of abuse over time-

increases the risk of cumulative negative effects (Ethier, Lemelin, & Lacharit, 2004; Jaffe &

Kohn Maikovich-Fong, 2011).

Disadvantage, “poly-violence” and educational outcomes

Adolescents exposed to more frequent “poly-violence” were less likely to be in the appropriate

grade for their age. This finding adds to the limited evidence on the negative effects of exposure

to violence on school progression and academic performance among adolescents in South

Africa (Pieterse, 2015; Sherr et al., 2016). However, most adolescents in the sample were high-

risk adolescents due to multiple economic disadvantages in the home, as well as further

disadvantages at school and in the community. For instance, most adolescents in the sample

experienced multiple stressors such as coming from impoverished families, living in low-income

communities and attending poorly-resourced schools. All these factors measuring socioeconomic

disadvantage have been associated with school delay and grade repetition in South Africa

(Branson et al., 2014; Lam et al., 2011; Spaull, 2015). Furthermore, the relationship between

poverty and lower levels of education during childhood has also been identified as a strong

predictor of violence perpetration and victimization in youth and adult life (WHO, 2002).

Our results showed that family poverty was not associated with exposure to violence.

However, the poverty measure in the current analysis needs to be interpreted as an indicator of

greater disadvantage in an already disadvantaged setting (all adolescents were highly at risk as

they came from socioeconomically disadvantaged communities). Further studies which compare

the exposure to violence of economically disadvantaged and economically non-disadvantaged

adolescents are needed in South Africa. These studies may shed further light on whether

incidences of exposure to multiple types of violence and its effects on educational outcomes

differ, depending on the socioeconomic status. Finally, our results suggest that the disadvantaged

communities in which participants lived may plausibly have contributed to the high rates of both

exposure to violence and school delay in the overall sample. However, a comparison group of

adolescents not exposed to violence was not included in the analysis.

Previous research on older adolescents has shown high perseverance, expectations and

motivation of disadvantaged South African youth to stay in school, regardless of their

socioeconomic circumstances (Bray et al., 2010; Strassburg et al., 2010; Ward et al., 2007). This

motivation to stay in school seems not to be associated with school completion and academic

achievement (Bray et al., 2010; Strassburg et al., 2010). Similarly, our disadvantaged sample

was also characterized by both high levels of academic motivation and school delay. Whilst

previous research on young children suggested that family violence negatively affected young

children’s academic motivation (Barbarin et al., 2001), here adolescents exposed to more

frequent “poly-violence” do not seem to have lower academic motivation in disadvantaged

contexts.

Several South African authors attempt to explain the high aspirations of disadvantaged

adolescents despite the lack of opportunities. Some think that youth’s aspirations can be

considered “unrealistic” within a context characterized by socioeconomic disadvantage due to an

inadequate grasping of practical constraints on such aspirations (Watson et al., 2010). Watson et

al. (2010) explain this further: ‘It is possible that children who live in disadvantaged

circumstances do not receive adequate exposure to the practical implications of their

occupational aspirations. As a result, they may not realistically understand the potential barriers

to actualizing these aspirations’. Certainly, these “unrealistic aspirations” would not be so

unrealistic if the majority of adolescents in South Africa were given the same opportunities; such

as having access to the same high-quality education that 25% of South African adolescents

receive (Spaull, 2015; Spaull & Kotze, 2015). This 25% are from higher socioeconomic

backgrounds who are not exposed to these structural barriers (Spaull, 2015). Others explain the

disassociation of dreams and opportunities within the current historical context of South Africa’s

young democracy, due to either a greater perception of social mobility or as a weapon against

misery (Bray, Gooskens, Kahn, Moses, & Seekings, 2010; Mzolo, 2013).

The relationship between adolescent age and their exposure to violence

Our results showed that age is associated with more frequent exposure to “poly-

violence” in adolescents from disadvantaged communities in South Africa. This finding seems

in line with evidence from high income countries where older children are more likely to be high

poly-victims, compared to low poly-victims (Finkelhor et al., 2007). However, our study focuses

on past-month exposure to violence compared to life time or past-year exposure to violence as

used in other studies (Finkelhor et al., 2007). Thus our finding suggests that adolescents’ risk to

higher levels of exposure to “poly-violence” is not only due to their longer lifespan, compared

with younger children. This risk of experiencing more frequent “poly-violence” in the past

month might be explained by a wider range of movement and more contact with peers;

adolescents tend to be less well monitored by caregivers due to increasing levels of autonomy

among older adolescents compared to younger ones.

Implications and Limitations

The study has several limitations. First, a purposive sampling approach was applied.

Adolescents were recruited from schools, community guides and social workers based on

knowledge of relational issues at home. Furthermore, families were selected using a screening

tool which asked about past arguments in the home. However, our recruitment and selection

procedures allowed for the inclusion of very vulnerable adolescents and their families. Second,

baseline characteristics showed that adolescents in the sample were simultaneously

disadvantaged in other ways (attending poorly-resourced schools, experiencing family poverty

etc.) Hence, it is likely that findings on the educational delay of adolescents may not be

explained by exposure to violence alone. Third, the six measures of exposure to violence used

adolescents’ self-reports. Self-report has often been criticized for being subject to recall bias.

Compared to other studies using past year or lifetime exposure to violence, this study used past

month report. The risk of recall bias was thus plausibly low, allowing us to identify very

vulnerable adolescents. Furthermore, the use of ACASI in the violence sections of the

questionnaire allowed us to reduce adolescents’ social desirability bias.

Fourth, period of reference for all the measures recording exposure to violence was

adapted to “in the past month” due to the RCT design and the intention to reduce participants’

recall bias. Although it is plausible that exposure to violence changes from one month the next

month, it is likely that certain practices leading to violence exposure in the home, such as harsh

parenting occurred as a consistent behaviour over the years (Meinck, Cluver, Orkin, et al., 2016).

Similarly, adolescents came from contexts characterized by deep-rooted community violence.

Thus, past-month exposure to school and community violence may be indicative of long-term

exposure to violence. Fifth, the time period of the present study (5 to 9 months) is not long

enough to understand the medium and long-term effects of adolescents’ exposure to violence on

education. Although all follow-up interviews were conducted during the next academic year,

compared to baseline interviews, we only measured a snapshot of adolescents’ experiences. It is

of course plausible that school delay had occurred over the years prior to our study; baseline

descriptive results showed that more than a third of all adolescents in the sample were already

enrolled in at least one grade lower than the one corresponding to their age. As previously

mentioned, it is also plausible that exposure to violence is likely to have happened more than a

month before our baseline data collection, which might have affected adolescents’ educational

outcomes before the beginning of the study. However, this study does not claim any causal

relationship between violence and school delay. In contrast, this investigation draws conclusions

about what may be possible based on a cross-sectional analysis. This is a first step to better

understanding the associations between violence, socioeconomic disadvantage and school delay.

A small proportion (n=31) of the sample did not report being exposed to multiple types

of violence including four adolescents who reported no violence exposure at all. However, the

analysis used in the present study yielded two distinct profiles, grouping all adolescents in the

sample based on the frequency of exposure to poly-violence. This is something that has occurred

in other studies. For instance, in Flannery, Wester, & Singer (2004), Gorman-smith et al. (2004)

and Yorohan (2011), children not exposed or exposed to one form of violence have constantly

been grouped with other children exposed to multiple types of violence in a low frequent

manner. These children have been considered as children exposed to ‘low’ levels of violence

(Flannery et al., 2004; Gorman-smith et al., 2004; Yorohan, 2011). However, our own study

differs on the methodology used, compared to these other studies. The main strength of using

LPA is that we have avoided using intuitive cut-off points to identify subgroups of adolescents

similarly exposed to violence and school delay.

There are at least three plausible reasons for our model not identifying a third group of

adolescents who had not been exposed to poly-violence nor violence. Firstly, this may be due to

having a very small representation of adolescents not exposed to poly-violence nor

violence. Secondly, in this socioeconomically disadvantaged context, adolescents not exposed to

poly-violence nor violence may experience similar school delay experiences compared to those

adolescents exposed to less-frequent poly-violence. Thirdly, the four adolescents who report not

being exposed to any type of violence were only doing so in the areas asked about in the past

month. These adolescents were living in the same communities as the vast majority of the

sample who reported the experience of poly-violence.  This raises a number of possibilities that

future research might investigate: 1. It may be that there is a small proportion (<1%) of

adolescents in these communities who experience none of the violence that this paper

investigates (even the indirect violence); 2. It may be that some adolescents in these

communities do not want to report the experience of violence to researchers; 3. It may be that

some adolescents don’t remember or don’t notice the violence in their communities; 4. It may be

that adolescents were exposed to poly-violence but not in the past month.  The extent to which

any of these are true would make for an interesting follow-up investigation. Further research

studies including more adolescents not necessarily exposed to more than one type of violence

(less at-risk adolescents) may shed further light on the different impacts of exposure to violence

on the educational outcomes of adolescents from socioeconomically disadvantaged communities

in South Africa.

Our findings suggest that adolescents from socioeconomically disadvantaged

communities are exposed to multiple stressors at home, in the school and in the community.

Poverty and exposure to violence affect adolescents’ lives and their education opportunities.

There is a need for a greater collaboration between different departments and stakeholders to

reduce domestic violence, abuse, school violence and crime in the community. Promoting

positive parenting practices while reducing financial stress at home may help reduce domestic

violence; addressing not only parent-adolescent relations but also socioeconomic structural risk

factors associated with abuse (Meinck, Cluver, & Boyes, 2013). Furthermore, more efforts are

needed to ensure that adolescents in South Africa feel safe within the school grounds. For

instance, low cost initiatives such as the Good School Toolkit study in Uganda can reduce

violence against children from school staff (Devries et al., 2015). Increased communication

among families, school teachers, police, community workers and social workers in the

communities is needed. Improving or developing networks for early referral systems between all

organizations working with children in order to prevent “poly-violence” is essential. More

research into resilience to family and community violence is much needed as well as further

studies looking at protective factors for educational outcomes in adolescents exposed to

violence. These may inform policy further and help tailor specific interventions for “poly-

victimized”, disadvantaged adolescents in South Africa.

References

Akmatov, M. K. (2011). Child abuse in 28 developing and transitional countries — results from the Multiple Indicator Cluster Surveys. International Journal of Epidemiology, (40), 219–227. http://doi.org/10.1093/ije/dyq168

Anonymous (2016). Details omitted for double-blind reviewing.

Anonymous et al (2018). Details omitted for double-blind reviewing.

Barbarin, O., Richter, L., & Wet, T. De. (2001). Exposure to Violence, Coping Resources, and Psychological Adjustment of South African Children. American Journal of Orthopsychiatry, 71(1), 16–25. http://doi.org/10.1037/0002-9432.71.1.16

Branson, N., Hofmeyr, C., & Lam, D. (2014). Progress through school and the determinants of school dropout in South Africa. Development Southern Africa, 31(1), 106–126. http://doi.org/10.1080/0376835X.2013.853610

Bray, R., Gooskens, I., Kahn, L., Moses, S., & Seekings, J. (2010). Growing up in the new South Africa. Childhood and adolescence in post-apartheid Cape Town. Cape Town: Human Sciences Research Council.

Bruwer, B., Govender, R., Bishop, M., Williams, D. R., Stein, D. J., & Seedat, S. (2014). Association between childhood adversities and long-term suicidality among South Africans from the results of the South African Stress and Health study : a cross-sectional study. BMJ, 4. http://doi.org/10.1136/bmjopen-2013-004644

Burton, P., & Leoschut, L. (2013a). School violence in South Africa: Results of the 2012 National School Violence Study. Monograph Series, No 12. Cape Town: Centre for Justice and Crime Prevention.

Burton, P., & Leoschut, L. (2013b). School Violence in South Africa Results of the 2012 National School Violence Study. Cape Town.

Charak, R., Byllesby, B. M., Roley, M. E., Meredith, A., Durham, T. A., Ross, J., … Elhai, J. D. (2016). Child Abuse & Neglect Latent classes of childhood poly-victimization and associations with suicidal behavior among adult trauma victims : Moderating role of anger. Child Abuse & Neglect, 62, 19–28. http://doi.org/10.1016/j.chiabu.2016.10.010

Clarke, K., Patalay, P., Allen, E., Knight, L., Naker, D., & Devries, K. (2016). Patterns and predictors of violence against children in Uganda : a latent class analysis. BMJ Open, 6, 1–10. http://doi.org/10.1136/bmjopen-2015-010443

Department of Basic Education. (2013). Education for All ( EFA ) 2013 Country Progress Report : South Africa. Pretoria.

Department of Basic Education. (2016). Schools Masterlist Data. Retrieved March 10, 2015, from http://www.education.gov.za/Programmes/EMIS/EMISDownloads.aspx

Devries, K. M., Knight, L., Child, J. C., Mirembe, A., Nakuti, J., Jones, R., … Allen, E. (2015). The Good School Toolkit for reducing physical violence from school staff to primary school students : a cluster-randomised controlled trial in Uganda. The Lancet Global Health, 3(385), e378–e386. http://doi.org/10.1016/S2214-109X(15)00060-1

Finkelhor, D., Ormrod, R. K., & Turner, H. A. (2007). Poly-victimization : A neglected component in child victimization. Child Abuse & Neglect, 31, 7–26.

http://doi.org/10.1016/j.chiabu.2006.06.008

Flannery, D. J., Wester, K. L., & Singer, M. I. (2004). Impact of exposure to violence in school on child and mental health and behavior. Journal of Community Psychology, 32(5), 559–573. http://doi.org/10.1002/jcop.20019

Frick, P. J. (1991). Alabama Parenting Questionnaire.

Gorman-smith, D., Henry, D. B., Tolan, P. H., Henry, D. B., Tolan, P. H., Gorman-smith, D., … Tolan, P. H. (2004). Exposure to Community Violence and Violence Perpetration : The Protective Effects of Family Functioning Exposure to Community Violence and Violence Perpetration : The Protective Effects of Family Functioning. Journal of Clinical Child and Adolescent Psychology, 33(3), 439–449. http://doi.org/10.1207/s15374424jccp3303

Hall, K., & Sambu, W. (2015). Income poverty, unemployment and social grants. Part 3 Children count - The numbers (Vol. South Afri).

Holt, M., Finkelhor, D., & Kantor, G. (2007). Multiple victimization experiences of urban elementary school students: Associations with psychosocial functioning and academic performance. Child Abuse & Neglect, 31, 503–515. http://doi.org/10.1016/j.chiabu.2006.12.006

Hox, J. (2010). Multilevel Anaysis Tecniques and Applications (2nd ed.). New York: Routledge.

Kitzmann, K. M., Gaylord, N. K., Holt, A. R., & Kenny, E. D. (2003). Child Witnesses to Domestic Violence : A Meta-Analytic Review. Journal of Consulting and Clinical Psychology, 71(2), 339–352. http://doi.org/10.1037/0022-006X.71.2.339

Lam, D., Ardington, C., & Leibbrandt, M. (2011). Schooling as a Lottery: racial differences in School Advancement in Urban South Africa. Journal of Developmental Economy, 95(2), 121–136. http://doi.org/10.1016/j.jdeveco.2010.05.005.Schooling

Lanza, S. T., Tan, X., & Bray, B. C. (2013). Latent class analysis with distal outcomes: a flexible model-based approach. Structural Equation Modeling : A Multidisciplinary Journal, 20(1), 1–26. http://doi.org/10.1080/10705511.2013.742377

Masitsa, M. G. (2011). Exploring safety in township secondary schools in the Free State province. South African Journal of Education, 31(1), 163–174.

Masten, A. S. (2014). Ordinary Magic: Resilience in Development. New York: The Guilford Press.

Masyn, K. (2013). Latent Class analysis and finite mixture modeling. In Oxford University Press (Ed.), The Oxford handbook of quantitative methods in psychology (Vol. 2) (pp. 551–611). New York.

Meinck, F., Cluver, L., & Boyes, M. (2013). Risk and Protective Factors for Physical and Emotional Abuse Victimisation amongst Vulnerable Children in South Africa. Child Abuse Review. http://doi.org/10.1002/car.2283

Meinck, F., Cluver, L. D., & Boyes, M. E. (2015). Household illness, poverty and physical and emotional child abuse victimisation : findings from South Africa’s first prospective cohort study. BMC Public Health, 15(444), 1–13. http://doi.org/10.1186/s12889-015-1792-4

Meinck, F., Cluver, L. D., Boyes, M. E., & Loening-voysey, H. (2016). Physical, emotional and sexual adolescent abuse victimisation in South Africa : prevalence, incidence, perpetrators and locations. Journal of Epidemial Community Health, 0, 1–7. http://doi.org/10.1136/jech-2015-

205860

Meinck, F., Cluver, L. D., Orkin, F. M., Kuo, C., Sharma, A. D., Hensels, I. S., & Sherr, L. (2016). Pathways From Family Disadvantage via Abusive Parenting and Caregiver Mental Health to Adolescent Health Risks in South Africa. Journal of Adolescent Health, 60(1), 57–64. http://doi.org/10.1016/j.jadohealth.2016.08.016

Meny-Gibert, S., & Russell, B. (2010). Enrolment, delays and completion in South African schools. Social Surveys Africa, 1.

Muthén, L. K., & Muthén, B. O. (2010). Mplus user’s guide (6th ed.). Los Angeles, CA: Muthén & Muthén.

Muthen, L., & Muthen, B. (2012). Mplus user’s guide (7th edition). (M. & Muthen, Ed.). Los Angeles (CA).

Mzolo, S. (2013). Fracturated Hope: celebrating 20 years of democracy amid poverty and despair. Cape Town: Trafford Publishing.

Nylund, K. L., & Muthén, B. O. (2007). Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling : A Monte Carlo Simulation Study. Structural Equation Modelling, 14(4), 535–569. http://doi.org/10.1080/10705510701575396

Petrenko, C., Friend, A., Garrido, E., Taussig, H., & Culhane, S. (2013). Does subtype matter? Assessing the effects of maltreatment on fucntioning in preadolescent youth in out-of-home care. Child Abuse & Neglect, 36(9), 633–644. http://doi.org/10.1016/j.chiabu.2012.07.001.Does

Pieterse, D. (2015). Childhood Maltreatment and Educational Outcomes: Evidence from South Africa Health Economics. Health Economics, 24(7), 876–894. http://doi.org/10.1002/hec.3065

Pillay, U., Roberts, B., & Rule, S. (2006). South African social attitudes: changing times, diverse voices. Cape Town: HSRC Press.

Ruchkin, V., Schwab-Stone, M., & Vermeiren, R. (2004). Social and Health Assessment (SAHA) Psychometric Development Summary. New Haven: Yale University.

Runyan, D. K., Dunne, M. P., Zolotor, A. J., Madrid, B., Jain, D., Gerbaka, B., … Youssef, R. M. (2009). The development and piloting of the ISPCAN Child Abuse Screening Tool-Parent version (ICAST-P). Child Abuse & Neglect, 33(11), 826–32. http://doi.org/10.1016/j.chiabu.2009.09.006

Sherr, L., Hensels, I. S., Skeen, S., Tomlinson, M., Roberts, K. J., & Macedo, A. (2016). Exposure to violence predicts poor educational outcomes in young children in South Africa and Malawi. International Health, (8), 36–43. http://doi.org/10.1093/inthealth/ihv070

Spaull, N. (2013). South Africa’s Education Crisis : The quality of education in South Africa 1994-2011. Centre for Development and enterprise (Vol. 27).

Spaull, N. (2015). Schooling in South Africa: How low quality education becomes a poverty trap. South African Child Gauge, (12), 34–41. Retrieved from http://nicspaull.com/research/

Spaull, N., & Kotze, J. (2015). Starting behind and staying behind in South Africa: The case of insurmountable learning deficits in mathematics. International Journal of Educational Development, (41), 13–24. http://doi.org/10.1016/j.ijedudev.2015.01.002

Sternberg, K. J., Baradaran, L. P., Abbott, C. B., Lamb, M. E., & Guterman, E. (2006). Type of

violence, age, and gender differences in the effects of family violence on children’s behavior problems : A mega-analysis. Developmental Review, 26, 89–112. http://doi.org/10.1016/j.dr.2005.12.001

Strassburg, S., Meny-Gibert, S., & Russell, B. (2010). Left unfinished. Social Surveys Africa, 2.

Strassburg, S., Meny-Gibert, S., & Russell, B. (2010). More than getting through the school gates. Social Surveys Africa, 3.

UNICEF. (2009). Child friendly schools programming. Global Evaluation Report. New York.

Van der Berg, S. (2008). How effective are poor schools? Poverty and educational outcomes in South Africa. Studies in Educational Evaluation, 34(3), 145–154. http://doi.org/10.1016/j.stueduc.2008.07.005

Van Wyk, C. (2015). An overview of key data sets in education in South Africa. South African Journal of Childhood Education, 5(2), 146–170. http://doi.org/10.4102/sajce.v5i2.394

Vranceanu, A., Hobfoll, S., & Johnson, R. (2007). Child multi-type maltreatment and associated depression and PTSD symptoms: The role of social support and stress. Child Abuse and Neglect, 31(1), 71–84. http://doi.org/10.1016/j.chiabu.2006.04.010

Ward, C., Artz, L., Burton, P., & Leoschut, L. (2015). The Optimus Study on Child Abuse, Violence and Neglect in South Africa.

Ward, C., Martin, E., Theron, C., & Distiller, G. (2007). Factors Affecting Resilience in Children Exposed to Violence. South African Journal of Psychology, 37(1), 165–187. http://doi.org/10.1177/008124630703700112

Watson, M., Mcmahon, M., Foxcroft, C., & Els, C. (2010). Occupational Aspirations of Low Socioeconomic Black South African. Journal of Career Development, 37(4), 717–734. http://doi.org/10.1177/0894845309359351

WHO. (2002). World report on violence and health. Geneva.

WHO. (2014). Chapter 3. Child abuse and neglect by parents and other caregivers. In World Health Organisation (Ed.), Global Status Report on Violence Prevention. Geneva.

Wright, G. (2008). Findings from the Indicators of Poverty and Social Exclusion Project: A Profile of Poverty using the Socially Perceived Necessities Approach. Key Report 7. Pretoria.

Yaw, A., Heaton, T. B., & Kalule-Sabiti, I. (2007). Living Arrangements in South Africa. In A. Yaw & T. B. Heaton (Eds.), Families and households in post-apartheid South Africa : Socio-demographic perspectives (pp. 43–60). Cape Town: HSRC Press.

Yorohan, R. (2011). The relationship between exposure to violence, acceptance of violence and engagement in violence: A study of Turkish adolescence [unpublished]. University. 2011. Istanbul Bilgi University.

Table 1 Sinovuyo Teen Study measures recording exposure to violence, academic motivation and school delayMeasures Sum Scale (#

Items) Items Items Response CodesExposure to violence

Domestic Violence between household members

ICASTDVT ICAST(2 items)

In the past month, how many days were there arguments with adults shouting in your home?In the past month, how many days were there arguments with adults hitting each other in your home?

0= Never1, 2 3, 4, 5, 6, 7 times8= 8 or more times

Adolescent physical abuse by caregivers

PHYABCPT ICAST-TRIAL + APQ(7 items)

In the past month, how often did an adult in your house ...Push, grab, or kick you?...Shake you?...Hit, beat, or spank you with a hand?...Hit, beat, or spank you with a belt, paddle, a stick or other object?

0= Never1, 2 3, 4, 5, 6, 7 times8= 8 or more times

Your caregiver spanks you with their hand when you have done something wrongYour caregiver slaps you when you have done something wrong.Your caregiver hits you with a belt, switch, or other object when you have done something wrong.

Original response values were multiplied by 2. New response codes: 0= Never; 2= Almost never; 4= Sometimes; 6= Often; 8= Always

Adolescent emotional abuse by caregivers

ICASTEMT ICAST-TRIAL(8 items)

In the past month how often did an adult …scream at you very loudly and aggressively?...Call you names, say mean things or swear at you?...Make you feel ashamed/embarrassed in front of other people in a way that made you feel bad?...Say that they wished you were dead or had never been born?...Threaten to leave you forever or abandon you?...Lock you out of the home for a long time?...Threaten to call evil spirits against you or hurt or kill you?...Refuse to speak to you because they were angry with you

0= Never1, 2 3, 4, 5, 6, 7 times8= 8 or more times

Perceived school violence

SCHUNST UNICEF (5 items)

I feel safe at school (reversed)I feel safe walking both to and from school (reversed)I sometimes don’t use the toilets at school because they are not safeThis school is being ruined by bulliesThis school is badly affected by crime and violence in the community

0= Definitely not true1= Mostly not true2= Mostly true3= Definitely true

Witnessing community violence

WCOMVIOT SAHA(5 items)

Someone else being threatenedSomeone else being mugged and his/ her your stuff stolenPeople fightingSomeone else being hit or harmedPeople being drunk or on drugs and being argumentative

0= Never1= Once or twice2= 3-5 times3= More than 5 times

Community violence victimization

VCOMVIOT SAHA(5 items)

Threatened by someone elseMugged and have your stuff stolenCaught up in a fightHit or harmed. With friends that were drunk or on drugs and argumentative

0= Never1= Once or twice2= 3-5 times3= More than 5 times

Educational Outcomes

(Low) Academic Motivation

LAMST3 SAHA(4 items)

It is important for me to do well in school this year (reversed)It is important to me to be considered clever by my fellow learners and my teachers (reversed)I try hard in school (reversed)I can’t wait to leave school

How strongly do you agree with the statements? Continuous response options from 0 to 9. Sum scale ranges from 0 to 36

School Delay SDECONT3School Delay(1 item)

-3= Three grades higher-2= Two grades higher-1= One grade higher0= Age appropriate grade1= One grade lower2= Two grades lower3= Three grades lower

Violence, school delay and academic motivation in South Africa

Table 2 Design effects and Intra Class Correlations (ICCs) of cross-classified levels of

nesting in Sinovuyo Teen Study (n=503)

　 School Delay Low Academic Motivation

　 Design effect

VarianceEstimate (p-value)

ICC Design effect

VarianceEstimate (p-value)

ICC

Community

1+ (12.575-

1)*0.152= 2.759

0.152 (0.03) 8.4%

1+ (12.575-

1)*0.254= 3.94

0.254 (0.201) 2.3%

School1+ (5.41-1)*0.072=

1.32

0.129 (0.072) 7.2%

1+ (5.41-1)*0.294=

2.3 0.294

(0.212)2.8%

School Average cluster size= 5.41Community Average cluster size = 12.575

33

Violence, school delay and academic motivation in South Africa

Table 3 Sample Characteristics at Baseline (n=503)

Mean (SD)/ n (%)DemographicsAdolescent Age (years) 13.71 (2.34)Female adolescent 208 (41.4%)Adolescent living in a rural community 405 (80.5%)Adolescent was an orphan 160 (31.8%)Xhosa as the main language spoken at home 502 (99.8%)Household socioeconomic characteristicsMore than 2 basic necessities missing in the past month 382 (75.9%)At least 2 days in the past week with not enough food at home 222 (44.1%)Living in a household where no one is working 350 (69.6%)Living in a household with no tap water 318 (63.2%)Schooling characteristicsEnrolled in school 503 (100%)Enrolled in at least one year below the appropriate grade in relation to age 128 (37.6%)Enrolled in at least two grades higher than the appropriate grade in relation to age

61 (13.1%)

Academic Motivation 28.35 (4.34)Attending secondary schools 214 (48.6%)Attending state schools 503 (100%)Attending schools in rural communities 335 (66.6%)Attending poorly-resourced schools1 430 (85.5%)Receiving free meals at school 484 (96.2%)Past-month exposure to violence2

Witnessed domestic violence between household members 257 (51.1%)Experienced physical abuse by caregivers 318 (63.2%)Experienced emotional abuse by caregivers 383 (76.1%)Felt unsafe in school 444 (88.3%)Experienced community violence (victimization) 163 (32.4%)Witnessed community violence 361 (71.8%)Not exposed to any form of violence 4 (0.8%)Exposed to “poly-violence” 472 (93.8%)Exposed to all six forms of violence 57 (11.3%)

1 Quintile 1-3 State Schools2 % of adolescents who replied different than never to at least one of the violence item questions or % of adolescents who replied mostly true or definitely true to at least one of the Perceived School Violence scale question

34

Violence, school delay and academic motivation in South Africa

Table 4 Statistical Criteria and profile sizes for latent profile analyses estimating 2-5 latent profiles

Number of profiles

N=503 (100%)

Log-L AIC BIC Entropy LMR test LMR, p value

2 -10126.363 20302.725 20408.240 0.934 432.8 0.0163 -9941.713 19951.425 20094.925 0.945 362.819 0.184 -9833.472 19752.944 19934.429 0.971 212.683 0.15 -9765.403 19634.807 19854.278 0.965 133.747 1

Profile 1 Profile 2 Profile 3 Profile 4 Profile 52 60 (11.9%) 443 (88.1%)3 59 (11.73%) 35 (6.96%) 409 (81.31%)4 32 (6.36%) 372 (73.96%) 28 (5.57%) 71 (14.12%)5 353 (70.18%) 33 (6.56%) 71 (14.12%) 28 (5.57%) 18 (3.58%)

35

Violence, school delay and academic motivation in South Africa

Table 5. Mean exposure to violence scores by profile membership

Exposure to Violence – Risk Indicators

Total sample: mean (SE)

LPA: profile comparison: mean (SE)

Profile 1More

frequent ‘poly-

violence’n=60

Profile 2Less

frequent ‘poly-

violence’n=443

p-value (Wald Test)

Domestic violence between household members 2.44

(0.16) 8.4 (1.13) 1.63 (0.21) <0.001

Adolescent physical abuse by caregivers 4.86

(0.28)16.77 (2.65)

6.01 (0.35) <0.001

Adolescent emotional abuse by caregivers 6.03

(0.36)21.58 (2.35)

3.93 (0.39) <0.001

Perceived school violence 6.19 (0.12) 9.2 (0.5) 8.86

(0.13) 0.542

Community violence – Victimization 0.72

(0.06) 1.19 (0.3) 0.66 (0.06) 0.063

Community violence – Witnessing 2.73 (0.12)

4.31 (0.56)

2.52 (0.14) 0.001

36

Violence, school delay and academic motivation in South Africa

Table 6 Characteristics of Adolescents exposed to more frequent ‘poly-violence’ and Adolescents exposed to less frequent ‘poly-violence’

Total Sample n=503

Profile 1More frequent ‘poly-violence’

n=60

Profile 2Less frequent ‘poly-violence’

n=443Demographics

Female (%) 208 (41.4%) 27 (45%) 181 (45%)Age (mean, years)1 13.71 (2.34) 14.5 (2.38) 13.55 (2.88)Family Poverty (mean, # of

basic necessities missing)3.28 (0.1) 3.63 (2.07) 3.23 (2.17)

Intervention Trial Arm (%) 252 (50.1%) 30 (50%) 221 (49.9%)Exposure to violence)

Domestic violence between household members (%)

257 (51.1%) 54 (90%) 203 (45.8%)

Adolescent physical abuse by caregivers (%)

318 (63.2%) 49 (81.7%) 269 (60.7%)

Adolescent emotional abuse by caregivers (%)

383 (76.1%) 60 (100%) 323 (72.9%)

Perceived school violence (%) 444 (88.3%) 55 (91.7%) 389 (87.8%)Community violence –

Victimization (%) 163 (32.4%) 26 (43.3%) 137 (30.9%)

Community violence – Witnessing (%) 361 (71.8%) 52 (86.7%) 309 (69.8%)

Not exposed to any form of violence (%) 4 (0.8%) 0 4 (0.9%)

Exposed to “poly-violence” (%) 472 (93.8%) 60 (100%) 412 (92.9%)

Exposed to all six forms of violence (%)2 57 (11.3%) 17 (28.3%) 40 (9%)

Educational outcomes (Low) Academic Motivation

(mean) 7.65 (0.19) 8.55 (0.74) 7.53 (0.23)

School Delay (mean) 1 0.03 (0.06) 0.52 (0.19) -0.04 (0.1)1Wald Test is statistically significant (p=0.007)2 Chi-square Test is statistically significant (p < 0.001)

37

Violence, school delay and academic motivation in South Africa

Table 7. Predictors of violence among adolescents from socioeconomically disadvantaged communities in South Africa

Factors associated with exposure to violence –Predictors or violence

Multinomial Logistic Regression: comparison of Profile 1 vs Profile 21

Standardized Estimate

(Beta)

SE p-value

Female (coded high) 0.22 0.31 0.470Age (years, older coded

higher)0.16 0.07 0.020

Family Poverty 0.08 0.07 0.230Membership of Trial Arm

(intervention)0.06 0.3 0.840

1Profile 2 is the reference category

38

Violence, school delay and academic motivation in South Africa

Figure 1 Patterns of exposure to violence among adolescents from socioeconomically disadvantaged communities in South Africa

39

* Episodes of violence

*